1 Faulty Generalizations

Unlike the scientist, the layman lacks techniques for testing generalizations. Laboratory instruments, trained technicians, controlled conditions -- these are all missing, There is no system for testing a generalization through all the multiplicity and complexity of the probable cases. Consider a very ordinary instance: "No human act is truly disinterested. Every man's generosity has somewhere a core of selfishness in it. Every man wants to get something when he gives something, if only self-approval." The "truly" takes care perhaps of the anonymous benefactor and similar alleged cases of disinterested generosity, by defining them away. But in all the deviousness of human relations, what means are available for searching out this core of selfishness? The constructions seem serviceably clear, but where can the evidence be found? Could all agree on what the evidence for confirming or falsifying the generalization might be? For instance, can any one suggest a test case that would be crucial for exposing self-interest, if present, in an act of pronounced generosity -- something like the crucial experiments of science?

These considerations have led skeptics to claim that in the business of daily life a principle of indifference operates. Many decisions are a gamble in which the dice can not be loaded, that is, in which there is not even a crooked way to anticipate the cast of fortune. And wagers, so to speak, are dictated by compulsive magical practices, mere habit, or wish-thinking. This is Richards' domain of random beliefs and hopeful guesses.

The facts fail to confirm the skeptical view of everyday thinking. Almost everybody who acts on principles and guides his present choices by his past experience meets in a thousand details of life with more successes than misfortunes. On the whole, intelligent people invest their money well, they order their human relationships pleasantly enough when they order them at all, they raise their children in mental and physical health. This cannot be all blind chance, balanced out by the bankrupts, the multiple-divorcees, and the heart-broken parents of delinquent juveniles. Disasters happen to very intelligent people, and sometimes "through their own fault." But by any common-sense standard of intelligence, they fall on them less often than on the uninformed and the unprincipled.

Information and principles, when expressed in language, take the form of generalizations. The language has a very limited number of ways to express generalizations. We usually have to say "all x is y" or "no x is y," or "most x is y" or "x somewhat resembles y" or some paraphrase of these. With increasing statistical and actuarial information, we are sometimes able to speak of the percentage of cases or the incidence of a trait. Sampling techniques guarantee to some extent an even distribution, so that the relatively small number of cases examined will provide reliable information about all the cases. Children from homes with an income i will become juvenile delinquents with a probability of p." "X percentage of married couples owning their own homes will divorce as compared with y percentage of those who pay rent."

Though the body of statistical information is growing daily, most generalizations on which laymen must base their understanding of the world they live in cannot claim any such precision of meaning. "Only psychopaths have no loyalties." "Brain washing and psychological torture prove what the Inquisition failed to demonstrate, that every man has his breaking point." "Child psychology will probably never develop an adequate safeguard against sibling jealousy."

These and thousands of similar statements seem entirely meaningful -- that is, they are true or false. They are sweeping; they are unprotected by sampling procedures or careful statistical sifting. Yet they or principles like them are indispensable in thinking about important problems.

The model of the statistical generalizations, nevertheless, throws light upon the nature of other generalizations. It is scarcely two generations since William James sent out the first known questionnaire (he wanted to find out if persons who had lost their legs felt that they had sensations in the missing members). In that brief period everybody has had many surprises as beliefs deriving from untested experience and "general knowledge" have tumbled over one after the other. The reflective man has learned a measure of caution.

He has also learned that when he says "all," "none," etc., he often means "almost all," "hardly any." In the realm of human behavior no (that is, hardly any) generalizations hold one hundred per cent. Even in medicine, a matter of chemistry and physiological structure, people occasionally recover from "incurable" diseases, walk around on fractures, and shoot themselves through the head without doing noticeable brain damage. As to common-sense beliefs in the psychological and cultural spheres, they remain tenuous to the point of being diaphanous.

Statistical findings suggest caution before placing confidence in universal statements. Sampling methods suggest extreme care in order to be representative. The refinement of questionnaire techniques suggests ways in which to ask questions so as not to influence the answers. The traditional "inductive" fallacies (see p. 13 ff) can be interpreted as infringements of rules justified by these procedures.

The positive criterion for testing a generalization is the literal truth of the proposition as stated. If the generalization claims "all," or "none," then one counter instance makes it false. "Finding the counter instance" is the great task of refutation in law courts, scientific journals, the learned controversies in philosophy, criticism, social science. It is also the most likely tool in the debates and discussions of every-day intelligent living.

Yet one must not pedantically carp at enthusiastic and sweeping statements which can be simply and plausibly recast without damage to a position. If a speaker says, "All power corrupts," he need not be rebuked. One can mentally add "by and large."

The two most common forms of faulty generalization are jumping to conclusions and selecting the cases. Let us take these up briefly and exemplify them.

(a) Hasty Generalization: "jumping to conclusions"

The fallacy of generalization from too few cases consists in drawing a general conclusion on the basis of an experience with particulars, which statistical science shows to be insufficient in view of the size of the unit examined, or, as pollsters say, the "population." Even where there is a lack of time or money to prepare a proper actuarial survey, there is no excuse for ignorance about the minimum size of the sample -- the man who talks to a few people in his office and to his neighbors and then makes a bet on a presidential election deserves to lose the bet.

Peter is eating Jane's Frozen Peas when he breaks a tooth on a pebble hidden among them. He tells all who will listen never to buy another Jane's product. Tired of hearing this complaint, Peter's wife points out that one bad experience with a box of peas does not constitute wide acquaintance with Jane's packaged products. It is not even a fair sample of Jane's Peas.
"Professor Ballast is insufferable, and I will have nothing to do with him. He struck his wife when she reminded him of his promise not to borrow more money." It is easier than in the previous example to feel sympathy with this general conclusion, which likewise stems from a smgle instance. Sometimes behavior is revealing because it is so far from the norm that a single instance indicates a bent of character. Although it is true that more instances would provide a better foundation for judgment, in a life situation one would hardly await numerous examples of unpardonable conduct before one felt secure in the conclusion "This man is insufferable."
Consider the following proposition: "The proverbs of the language preserve generalizations that have stood the test of time. They embody the experience and wisdom of the race." Several things are wrong here, but the present concern is primarily with insufficient cases. The speaker presumably refers to such expressions as "A stitch in time saves nine," and "A bad penny always returns." But proverbs often contradict each other: "Look before you leap," "He who hesitates is lost." Or, "Two's company, three's a crowd," "The more the merrier." Or, "Absence makes the heart grow fonder," "Out of sight, out of mind." Counter instances have shown that the generalization about "generalizations that have stood the test of time" is false.
A student of contemporary literature writes, "So-called science fiction is a waste of time. I have waded through nine novels and many short stories by the better known writers in science fiction, and I find that their plots are feeble, their characterizations implausible and the science in the fiction is fantastic and largely irrelevant to the action." In answer, the editor of a science fiction magazine points to two examples of well constructed novels in the field, which have had a deserved success. The editor's argument, citing counter instances, is legitimate in testing the critic's unfavorable generalization. The generalization "All science fiction is bad writing" is exploded by pointing to one good story or novel. In the everyday use of language, people are not so strict. The critic here need not be interpreted as saying that all science fiction writing is bad. He is establishing a critical judgment and describing the genre as a whole. It is clearly a "waste of time," as he says, to read for hundreds of hours in order to find one or two good stories.

(b) Unrepresentative Generalization

Though people in the ordinary case will not be able to sample a widespread population, in the manner of a Gallup poll, knowledge of sampling techniques ought to make them shy of regional, occupational, and other differences. Polling experience abundantly demonstrates that even a very extensive acquaintance with particulars does not warrant a universal conclusion unless the particulars enjoy a representative dispersion. If a minister should poll every member in a large metropolitan congregation of a certain church on, say, capital punishment, this will not enable him to predict with any assurance the division of opinion country-wide, even in the same religious denomination.

The extreme instance of unrepresentative dispersion of cases is the misrepresented selection. Many people have an impulse to suppress the counter instances: consciously or unconsciously they vitiate the evidence by selecting only a part of it. They look exclusively for evidence that will support an idea, with the result that they overlook the opposed evidence. The selection may take a more subtle form: no actual case is suppressed, but the sample is chosen to avoid damaging cases. Side-walk interviews held on prominent shopping streets are often of this sort. An investigation of the effectiveness of TV advertising could be made to yield decidedly different results, depending on whether the survey was made near a college campus or in an underprivileged part of town. Much cited in this connection is the Literary Digest poll which predicted the election of Landon over Roosevelt in 1936. An egregious error resulted from an inadequate sampling technique: in 1936, apparently, millions of Democrats could not afford a telephone, or had no time for questionnaires, or had not formed the middle-class "clerical habit" of answering them.
A famous historian writes: "Military coalitions usually fall apart before their objective is attained. So it was with the coalitions which opposed Louis XIV, Frederick the Great, and Napoleon I." The writer is suppressing, one hopes unconsciously, evidence against his generalization. The academese "usually" does not let the good scholar escape the charge of giving a false generalization, unless he is prepared to show that an adequate number of instances justify the assertion "usually." To hedge against contradiction by employing without evidence "usually" or any like words is to make unjustified assertions. In this case, the exceptions to the statement are many -- for instance, both Louis XIV and Napoleon I were ultimately defeated by coalitions that remained mtact against them. Incidentally, on August 31, 1944, Hitler stated in a conference of his generals, "All the coalitions in history have disintegrated sooner or later. The only thing is to wait for the right moment, no matter how hard it is." Thus, in the hope that the allies would fall out among themselves before his defeat, Hitler justified prolonging World War II until battle raged in the streets of Berlin. One may conjecture that Hitler's generals would hardly have dared to subject his generalization to an impartial discussion, and that all present had reason to indulge in rationalization (see #25).
Professor Peter is talking to a colleague. "Engineering students generally study hard. I have three engineers taking my sophomore course, and they're all hard workers." The professor should be allowed some leeway in expressing ideas in a casual conversation. At least, he is trying to cite confirmatory evidence for his opinions. The danger is that, having advanced an idea, he will look for cases from now on that confirm it, rather than examining a representative group of instances. Moreover, the number here is so far too small to permit any but the most hasty generalization.
Overheard on a bus, "That Maud is illiterate. She says, 'I ain't got none' for 'I haven't got none.' " That the correction is also "illiterate" is not the point of this example. The word "ain't" is a sort of shibboleth for the semi-educated; those who use it brand themselves as linguistically hopeless. For the better educated there are many other crucial cases, such as the use of "none" for "any." The interesting thing is the actual presence of such definite criteria. Consider the example of the professor who struck his wife (see pp. 13-14). On the basis of this one act, people are ready to condemn Professor Ballast as being socially hopeless. No one asks if the act is representative. The question "Out of a thousand hours with his wife, how many does Professor Ballast spend striking her?" seems ridiculous. Similarly, people don't require to know how often a speaker says "ain't got" for "haven't got." One occurrence of certain traits is felt to be "representative" enough for the purpose of judgment. But care must be exercised in the extent of the judgment. A speaker who says "ain't" may expose himself as belonging to a certain linguistic group or as employing a certain dialect, but it does not follow that what he says in his idiom may not make excellent sense, or that his speech may not evidence candor and sense. As to men who strike their wives, whatever the provocation, they reveal a serious character fault, and they will be judged accordingly by their associates. Again, it does not follow that their character is all fault. Many men with dangerous tempers have been known to be generally kind, considerate, gentle. The problem, then, in these cases is not that the generalization is hasty or unrepresentative, but that the judgments based upon it may go beyond it. From "bad grammar" to "stupid speaker" is an easy but snobbish step. The leap from "bad temper" to "bad man" is unwarranted and uncharitable.